Shape guidance improves 3D X-ray reconstruction with few views
Shape-guided Gaussian Splatting for Sparse-View X-ray 3D Reconstruction
Computer Vision and Pattern Recognition
Summary
It is hard to create detailed 3D X-ray images when only a few X-ray pictures are taken, because the computer guesses many unknown details. The authors improved this by using prior knowledge about typical body shapes to guide the reconstruction. This makes the 3D images more accurate and realistic, even with as few as five X-ray views. Their method beats previous techniques by making the results more consistent with human anatomy.
What this means in practice
- •For medical imaging teams: Produce higher-quality 3D X-ray images from limited viewpoints, reducing patient radiation exposure during scans.
- •For medical device developers: Create compact or faster X-ray imaging devices that need fewer captures without losing accuracy in 3D reconstruction.$Commercial implications: Enables development of advanced X-ray scanners with reduced radiation dose marketed to hospitals seeking safer imaging solutions.
Authors
Pranav Poudel, Florence Dell'Aniello Picard, Nairouz Shehata, Frédéric Lavoie, Herve Lombaert
Abstract
Sparse-view X-ray 3D reconstruction is essential for reducing radiation exposure, but recovering a density field from a handful of X-ray projections is severely ill-posed. Recently, 3D Gaussian Splatting has achieved state-of-the-art performance in sparse-view reconstruction by representing the volume using explicit, optimized primitives, but it requires dozens of projected views. With fewer views, reconstruction quality degrades severely since the explicit primitives are optimized freely without any anatomical information. Anatomical structures, in contrast, share similar geometry and density across a population. Their variations are bounded within a limited range that statistical shape models can capture. This paper proposes a shape-guided Gaussian splatting framework for sparse-view X-ray 3D reconstructions. Our contribution lies in driving Gaussian positions toward anatomically valid configurations, alongside atlas-based density regularization. Our method ensures anatomically consistent reconstruction and improves PSNR by 2.83 dB over a state-of-the-art Gaussian splatting baseline with as few as 5 views. Code Available: https://github.com/polyshape-lab/ShapeGuidedGaussian